A Division Method in Combustion Mode to Improve the Forecast Accuracy of the Cylinder Pressure Based on the Clustering Method

Pressure prediction technology is to predict in-Cylinder pressure through those state variables that are easy to observe and related to Cylinder pressure, which could overcome the high cost and some other shortcomings of the transducer. However, the existing cylinder pressure prediction methods have the problem that the prediction effect is not satisfactory when facing the pressure curve with complex shape under a wide range of working conditions. Therefore, in order to improve the performance of the Cylinder pressure prediction algorithm, we propose a combustion mode division method to construct prediction algorithms for cylinder pressure under different combustion modes respectively. This method is capable of classifying combustion patterns, which uses an unsupervised learning approach, that does not require human labeling or human definition of combustion patterns. Three kinds of cylinder pressure prediction algorithms are used to examine the effect of the proposed method on the cylinder pressure estimation effect. The results show that the pressure curves in the same combustion mode divided by this method are similar in shape and working conditions, and there are obvious differences between different combustion modes. Timing and load have a greater impact on these combustion mode divisions. The prediction performance of the three algorithms is improved after the combustion pattern recognition algorithm is added.